Skip to content

Commit

Permalink
some fixes and optimizations
Browse files Browse the repository at this point in the history
  • Loading branch information
@ivan-vasilev
ivan-vasilev committed Apr 17, 2014
1 parent a731467 commit 1614d86
Show file tree
Hide file tree
Showing 17 changed files with 169 additions and 122 deletions.
7 changes: 2 additions & 5 deletions nn-core/src/main/java/com/github/neuralnetworks/architecture/ConnectionFactory.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -5,6 +5,7 @@
import java.util.Arrays;
import java.util.List;

import com.github.neuralnetworks.util.Environment;
import com.github.neuralnetworks.util.Matrix;
import com.github.neuralnetworks.util.Tensor;
import com.github.neuralnetworks.util.TensorFactory;
Expand All @@ -21,11 +22,7 @@ public class ConnectionFactory implements Serializable {

public ConnectionFactory() {
super();
}

public ConnectionFactory(boolean useSharedMemory) {
super();
if (useSharedMemory) {
if (Environment.getInstance().getUseWeightsSharedMemory()) {
this.connections = new ArrayList<>();
this.sharedWeights = new float[0];
}
Expand Down
88 changes: 44 additions & 44 deletions nn-core/src/main/java/com/github/neuralnetworks/architecture/types/NNFactory.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -46,7 +46,7 @@ public class NNFactory {
* @param useSharedMemory - whether all network weights should be in a single array
* @return neural network
*/
public static NeuralNetworkImpl convNN(int[][] layers, boolean addBias, boolean useSharedMemory) {
public static NeuralNetworkImpl convNN(int[][] layers, boolean addBias) {
if (layers.length <= 1) {
throw new IllegalArgumentException("more than one layer is required");
}
Expand All @@ -56,7 +56,7 @@ public static NeuralNetworkImpl convNN(int[][] layers, boolean addBias, boolean
}

NeuralNetworkImpl result = new NeuralNetworkImpl();
ConnectionFactory cf = new ConnectionFactory(useSharedMemory);
ConnectionFactory cf = new ConnectionFactory();

Layer prev = null;
int prevUnitCount = layers[0][0] * layers[0][1] * layers[0][2];
Expand Down Expand Up @@ -131,9 +131,9 @@ public static NeuralNetworkImpl convNN(int[][] layers, boolean addBias, boolean
* @param useSharedMemory - whether all network weights will be part of single array
* @return
*/
public static NeuralNetworkImpl mlp(int[] layers, boolean addBias, boolean useSharedMemory) {
public static NeuralNetworkImpl mlp(int[] layers, boolean addBias) {
NeuralNetworkImpl result = new NeuralNetworkImpl();
mlp(result, new ConnectionFactory(useSharedMemory), layers, addBias);
mlp(result, new ConnectionFactory(), layers, addBias);
return result;
}

Expand Down Expand Up @@ -301,63 +301,63 @@ public static void lcStochasticPooling(NeuralNetworkImpl nn) {
}
}

public static NeuralNetworkImpl mlpSigmoid(int[] layers, boolean addBias, boolean useSharedMemory) {
NeuralNetworkImpl result = mlp(layers, addBias, useSharedMemory);
public static NeuralNetworkImpl mlpSigmoid(int[] layers, boolean addBias) {
NeuralNetworkImpl result = mlp(layers, addBias);
result.setLayerCalculator(lcSigmoid(result, null));
return result;
}

public static NeuralNetworkImpl mlpSoftRelu(int[] layers, boolean addBias, boolean useSharedMemory, ConnectionCalculator outputCC) {
NeuralNetworkImpl result = mlp(layers, addBias, useSharedMemory);
public static NeuralNetworkImpl mlpSoftRelu(int[] layers, boolean addBias, ConnectionCalculator outputCC) {
NeuralNetworkImpl result = mlp(layers, addBias);
result.setLayerCalculator(lcSoftRelu(result, outputCC));
return result;
}

public static NeuralNetworkImpl mlpRelu(int[] layers, boolean addBias, boolean useSharedMemory, ConnectionCalculator outputCC) {
NeuralNetworkImpl result = mlp(layers, addBias, useSharedMemory);
public static NeuralNetworkImpl mlpRelu(int[] layers, boolean addBias, ConnectionCalculator outputCC) {
NeuralNetworkImpl result = mlp(layers, addBias);
result.setLayerCalculator(lcRelu(result, outputCC));
return result;
}

public static NeuralNetworkImpl mlpTanh(int[] layers, boolean addBias, boolean useSharedMemory, ConnectionCalculator outputCC) {
NeuralNetworkImpl result = mlp(layers, addBias, useSharedMemory);
public static NeuralNetworkImpl mlpTanh(int[] layers, boolean addBias, ConnectionCalculator outputCC) {
NeuralNetworkImpl result = mlp(layers, addBias);
result.setLayerCalculator(lcTanh(result, outputCC));
return result;
}

public static Autoencoder autoencoder(int visibleCount, int hiddenCount, boolean addBias, boolean useSharedMemory) {
public static Autoencoder autoencoder(int visibleCount, int hiddenCount, boolean addBias) {
Autoencoder result = new Autoencoder();
mlp(result, new ConnectionFactory(useSharedMemory), new int[] {visibleCount, hiddenCount, visibleCount}, addBias);
mlp(result, new ConnectionFactory(), new int[] {visibleCount, hiddenCount, visibleCount}, addBias);
return result;
}

public static Autoencoder autoencoderSigmoid(int visibleCount, int hiddenCount, boolean addBias, boolean useSharedMemory) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias, useSharedMemory);
public static Autoencoder autoencoderSigmoid(int visibleCount, int hiddenCount, boolean addBias) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias);
ae.setLayerCalculator(lcSigmoid(ae, null));
return ae;
}

public static Autoencoder autoencoderSoftReLU(int visibleCount, int hiddenCount, boolean addBias, boolean useSharedMemory, ConnectionCalculator outputCC) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias, useSharedMemory);
public static Autoencoder autoencoderSoftReLU(int visibleCount, int hiddenCount, boolean addBias, ConnectionCalculator outputCC) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias);
ae.setLayerCalculator(lcSoftRelu(ae, outputCC));
return ae;
}

public static Autoencoder autoencoderReLU(int visibleCount, int hiddenCount, boolean addBias, boolean useSharedMemory, ConnectionCalculator outputCC) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias, useSharedMemory);
public static Autoencoder autoencoderReLU(int visibleCount, int hiddenCount, boolean addBias, ConnectionCalculator outputCC) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias);
ae.setLayerCalculator(lcRelu(ae, outputCC));
return ae;
}

public static Autoencoder autoencoderTanh(int visibleCount, int hiddenCount, boolean addBias, boolean useSharedMemory, ConnectionCalculator outputCC) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias, useSharedMemory);
public static Autoencoder autoencoderTanh(int visibleCount, int hiddenCount, boolean addBias, ConnectionCalculator outputCC) {
Autoencoder ae = autoencoder(visibleCount, hiddenCount, addBias);
ae.setLayerCalculator(lcTanh(ae, outputCC));
return ae;
}

public static RBM rbm(int visibleCount, int hiddenCount, boolean addBias, boolean useSharedMemory) {
public static RBM rbm(int visibleCount, int hiddenCount, boolean addBias) {
RBM result = new RBM();
ConnectionFactory cf = new ConnectionFactory(useSharedMemory);
ConnectionFactory cf = new ConnectionFactory();
result.addConnections(cf.fullyConnected(new Layer(), new Layer(), visibleCount, hiddenCount));

if (addBias) {
Expand Down Expand Up @@ -406,13 +406,13 @@ public static RBMLayerCalculator rbmTanhTanh(RBM rbm, int batchSize) {
return new RBMLayerCalculator(rbm, batchSize, new AparapiTanh(), new AparapiTanh(), new AparapiTanh());
}

public static DBN dbn(int[] layers, boolean addBias, boolean useSharedMemory) {
public static DBN dbn(int[] layers, boolean addBias) {
if (layers.length <= 1) {
throw new IllegalArgumentException("more than one layer is required");
}

DBN result = new DBN();
ConnectionFactory cf = new ConnectionFactory(useSharedMemory);
ConnectionFactory cf = new ConnectionFactory();
result.addLayer(new Layer());
for (int i = 1; i < layers.length; i++) {
RBM rbm = new RBM();
Expand All @@ -429,36 +429,36 @@ public static DBN dbn(int[] layers, boolean addBias, boolean useSharedMemory) {
return result;
}

public static DBN dbnSigmoid(int[] layers, boolean addBias, boolean useSharedMemory) {
DBN result = dbn(layers, addBias, useSharedMemory);
public static DBN dbnSigmoid(int[] layers, boolean addBias) {
DBN result = dbn(layers, addBias);
result.setLayerCalculator(lcSigmoid(result, null));
return result;
}

public static DBN dbnSoftReLU(int[] layers, boolean addBias, boolean useSharedMemory) {
DBN result = dbn(layers, addBias, useSharedMemory);
public static DBN dbnSoftReLU(int[] layers, boolean addBias) {
DBN result = dbn(layers, addBias);
result.setLayerCalculator(lcSoftRelu(result, null));
return result;
}

public static DBN dbnReLU(int[] layers, boolean addBias, boolean useSharedMemory) {
DBN result = dbn(layers, addBias, useSharedMemory);
public static DBN dbnReLU(int[] layers, boolean addBias) {
DBN result = dbn(layers, addBias);
result.setLayerCalculator(lcRelu(result, null));
return result;
}

public static DBN dbnTanh(int[] layers, boolean addBias, boolean useSharedMemory) {
DBN result = dbn(layers, addBias, useSharedMemory);
public static DBN dbnTanh(int[] layers, boolean addBias) {
DBN result = dbn(layers, addBias);
result.setLayerCalculator(lcTanh(result, null));
return result;
}

public static StackedAutoencoder sae(int[] layers, boolean addBias, boolean useSharedMemory) {
public static StackedAutoencoder sae(int[] layers, boolean addBias) {
if (layers == null || layers.length <= 1) {
throw new IllegalArgumentException("more than one layer is required");
}

ConnectionFactory cf = new ConnectionFactory(useSharedMemory);
ConnectionFactory cf = new ConnectionFactory();
StackedAutoencoder result = new StackedAutoencoder(new Layer());
for (int i = 1; i < layers.length; i++) {
Autoencoder ae = new Autoencoder();
Expand All @@ -472,26 +472,26 @@ public static StackedAutoencoder sae(int[] layers, boolean addBias, boolean useS
return result;
}

public static StackedAutoencoder saeSigmoid(int[] layers, boolean addBias, boolean useSharedMemory) {
StackedAutoencoder sae = sae(layers, addBias, useSharedMemory);
public static StackedAutoencoder saeSigmoid(int[] layers, boolean addBias) {
StackedAutoencoder sae = sae(layers, addBias);
sae.setLayerCalculator(lcSigmoid(sae, null));
return sae;
}

public static StackedAutoencoder saeSoftReLU(int[] layers, int hiddenCount, boolean addBias, boolean useSharedMemory) {
StackedAutoencoder sae = sae(layers, addBias, useSharedMemory);
public static StackedAutoencoder saeSoftReLU(int[] layers, int hiddenCount, boolean addBias) {
StackedAutoencoder sae = sae(layers, addBias);
sae.setLayerCalculator(lcSoftRelu(sae, null));
return sae;
}

public static StackedAutoencoder saeReLU(int[] layers, int hiddenCount, boolean addBias, boolean useSharedMemory) {
StackedAutoencoder sae = sae(layers, addBias, useSharedMemory);
public static StackedAutoencoder saeReLU(int[] layers, int hiddenCount, boolean addBias) {
StackedAutoencoder sae = sae(layers, addBias);
sae.setLayerCalculator(lcRelu(sae, null));
return sae;
}

public static StackedAutoencoder saeTanh(int[] layers, int hiddenCount, boolean addBias, boolean useSharedMemory) {
StackedAutoencoder sae = sae(layers, addBias, useSharedMemory);
public static StackedAutoencoder saeTanh(int[] layers, int hiddenCount, boolean addBias) {
StackedAutoencoder sae = sae(layers, addBias);
sae.setLayerCalculator(lcTanh(sae, null));
return sae;
}
Expand Down
2 changes: 1 addition & 1 deletion nn-core/src/main/java/com/github/neuralnetworks/calculation/RBMLayerCalculator.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -32,7 +32,7 @@ public RBMLayerCalculator(RBM rbm, int miniBatchSize, ConnectionCalculator posPh
this.posPhaseCC = posPhaseCC;
this.negPhaseVisibleToHiddenCC = negPhaseVisibleToHiddenCC;
this.negPhaseHiddenToVisibleCC = negPhaseHiddenToVisibleCC;
this.posPhaseVP = TensorFactory.tensorProvider(rbm, miniBatchSize, Environment.getInstance().getUseSharedMemory());
this.posPhaseVP = TensorFactory.tensorProvider(rbm, miniBatchSize, Environment.getInstance().getUseDataSharedMemory());
this.negPhaseVP = TensorFactory.tensorProvider(posPhaseVP, rbm);
}

Expand Down
2 changes: 1 addition & 1 deletion ...re/src/main/java/com/github/neuralnetworks/training/DeepTrainerTrainingInputProvider.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -30,7 +30,7 @@ public DeepTrainerTrainingInputProvider(TrainingInputProvider inputProvider, DNN
this.dnn = dnn;
this.currentNN = currentNN;
this.calculatedLayers = new HashSet<>();
this.layerResults = TensorFactory.tensorProvider(batchSize, Environment.getInstance().getUseSharedMemory(), dnn, currentNN);
this.layerResults = TensorFactory.tensorProvider(batchSize, Environment.getInstance().getUseDataSharedMemory(), dnn, currentNN);
this.inputDataBase = new TrainingInputDataImpl(layerResults.get(dnn.getInputLayer()));
}

Expand Down
2 changes: 1 addition & 1 deletion nn-core/src/main/java/com/github/neuralnetworks/training/Trainer.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -65,7 +65,7 @@ public void test() {
triggerEvent(new TestingStartedEvent(this));

Set<Layer> calculatedLayers = new UniqueList<>();
ValuesProvider results = TensorFactory.tensorProvider(n, getTestBatchSize(), Environment.getInstance().getUseSharedMemory());
ValuesProvider results = TensorFactory.tensorProvider(n, getTestBatchSize(), Environment.getInstance().getUseDataSharedMemory());

OutputError oe = getOutputError();
if (oe != null) {
Expand Down
4 changes: 2 additions & 2 deletions .../main/java/com/github/neuralnetworks/training/backpropagation/BackPropagationTrainer.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -31,9 +31,9 @@ public class BackPropagationTrainer<N extends NeuralNetwork> extends OneStepTrai

public BackPropagationTrainer(Properties properties) {
super(properties);
activations = TensorFactory.tensorProvider(getNeuralNetwork(), getTrainingBatchSize(), Environment.getInstance().getUseSharedMemory());
activations = TensorFactory.tensorProvider(getNeuralNetwork(), getTrainingBatchSize(), Environment.getInstance().getUseDataSharedMemory());
activations.add(getProperties().getParameter(Constants.OUTPUT_ERROR_DERIVATIVE), activations.get(getNeuralNetwork().getOutputLayer()).getDimensions());
backpropagation = TensorFactory.tensorProvider(getNeuralNetwork(), getTrainingBatchSize(), Environment.getInstance().getUseSharedMemory());
backpropagation = TensorFactory.tensorProvider(getNeuralNetwork(), getTrainingBatchSize(), Environment.getInstance().getUseDataSharedMemory());
}

/* (non-Javadoc)
Expand Down
2 changes: 1 addition & 1 deletion nn-core/src/main/java/com/github/neuralnetworks/training/events/EarlyStoppingListener.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -66,7 +66,7 @@ public void handleEvent(TrainingEvent event) {

ValuesProvider vp = mbe.getResults();
if (vp == null) {
vp = TensorFactory.tensorProvider(n, 1, Environment.getInstance().getUseSharedMemory());
vp = TensorFactory.tensorProvider(n, 1, Environment.getInstance().getUseDataSharedMemory());
}
if (vp.get(outputError) == null) {
vp.add(outputError, vp.get(n.getInputLayer()).getDimensions());
Expand Down
28 changes: 14 additions & 14 deletions nn-core/src/main/java/com/github/neuralnetworks/util/Environment.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -20,19 +20,19 @@ public class Environment {
private KernelExecutionStrategy executionStrategy;

/**
* is debug
* Shared memory for calculations
*/
private boolean debug;
private boolean useDataSharedMemory;

/**
* Shared memory
* Shared memory for neural network connection weights
*/
private boolean useSharedMemory;
private boolean useWeightsSharedMemory;

private Environment() {
executionStrategy = new DefaultKernelExecution();
debug = true;
useSharedMemory = true;
useDataSharedMemory = true;
useWeightsSharedMemory = false;
}

public KernelExecutionStrategy getExecutionStrategy() {
Expand Down Expand Up @@ -62,19 +62,19 @@ public static Environment getInstance() {
return singleton;
}

public boolean isDebug() {
return debug;
public boolean getUseDataSharedMemory() {
return useDataSharedMemory;
}

public void setDebug(boolean debug) {
this.debug = debug;
public void setUseDataSharedMemory(boolean useDataSharedMemory) {
this.useDataSharedMemory = useDataSharedMemory;
}

public boolean getUseSharedMemory() {
return useSharedMemory;
public boolean getUseWeightsSharedMemory() {
return useWeightsSharedMemory;
}

public void setUseSharedMemory(boolean useSharedMemory) {
this.useSharedMemory = useSharedMemory;
public void setUseWeightsSharedMemory(boolean useWeightsSharedMemory) {
this.useWeightsSharedMemory = useWeightsSharedMemory;
}
}
3 changes: 2 additions & 1 deletion nn-core/src/test/java/com/github/neuralnetworks/test/AETest.java
View file
Original file line number Diff line number Diff line change
Expand Up @@ -27,7 +27,8 @@ public void testAEBackpropagation() {
Environment.getInstance().setExecutionMode(EXECUTION_MODE.CPU);

// autoencoder with 6 input/output and 2 hidden units
Autoencoder ae = NNFactory.autoencoderSigmoid(6, 2, true, true);
Environment.getInstance().setUseWeightsSharedMemory(true);
Autoencoder ae = NNFactory.autoencoderSigmoid(6, 2, true);

// We'll use a simple dataset of symptoms of a flu illness. There are 6
// input features and the first three are symptoms of the illness - for
Expand Down
Loading

0 comments on commit 1614d86

Please sign in to comment.